一种任意时间片长度对CPU上下文切换代价影响消除的新方法
发布时间：2018-07-16 14:06
【摘要】：分析并量化了操作系统任务调度时上下文切换对CPU性能影响,得出了任务上下文切换代价对CPU效率影响关系.在此影响关系的基础上,提出了一种可行的Hyper-Scheduling方法.此方法通过在CPU内部设置一条特殊的任务切换专用流水线数据通路,并在通用寄存器堆上寄生一个相同的寄存器堆(影子寄存器堆)和监视访存状态的寄生逻辑,来监视各硬件资源运行情况,实现在CPU硬件资源闲置状态时进行预先数据准备或任务保存.当任务切换时将这条特殊数据通路与CPU主数据通路流水线进行交换,使上下文切换时间可趋近于零,任务切换代价被消除,上下文频繁切换或时间片长度缩短至近于零的情况下,CPU效率仍能够保持性能最大化.
[Abstract]:The influence of context switching on CPU performance is analyzed and quantified, and the relationship between task context switching cost and CPU efficiency is obtained. Based on the relationship, a feasible Hyper-Scheduling method is proposed. In this method, a special task switching special pipeline data path is set up inside the CPU, and a common register file (shadow register file) is parasitized on the general register file and the parasitic logic of monitoring the memory access state is monitored. To monitor the running situation of each hardware resource, the CPU hardware resource can be prepared or saved in advance when the CPU hardware resource is idle. When the task switches, the special data path is exchanged with the CPU main data path pipeline, so that the context switching time can approach zero, and the task switching cost is eliminated. The CPU efficiency can be maximized when the context is switched frequently or the time slice length is shortened to near zero.
【作者单位】： 华东师范大学可信物联网产学研联合研发中心;华东师范大学软件学院;
【分类号】：TP332
本文编号：2126629
[Abstract]:The influence of context switching on CPU performance is analyzed and quantified, and the relationship between task context switching cost and CPU efficiency is obtained. Based on the relationship, a feasible Hyper-Scheduling method is proposed. In this method, a special task switching special pipeline data path is set up inside the CPU, and a common register file (shadow register file) is parasitized on the general register file and the parasitic logic of monitoring the memory access state is monitored. To monitor the running situation of each hardware resource, the CPU hardware resource can be prepared or saved in advance when the CPU hardware resource is idle. When the task switches, the special data path is exchanged with the CPU main data path pipeline, so that the context switching time can approach zero, and the task switching cost is eliminated. The CPU efficiency can be maximized when the context is switched frequently or the time slice length is shortened to near zero.
【作者单位】： 华东师范大学可信物联网产学研联合研发中心;华东师范大学软件学院;
【分类号】：TP332
【相似文献】
相关期刊论文 前3条
1 吴贞海;刘福岩;;基于快速上下文切换扩展的快速地址空间切换[J];计算机工程;2010年10期
2 刘月吉;张盛兵;黄嵩人;;一种DSP的快速上下文切换机制[J];计算机应用研究;2012年01期
3 崔光佐,李晓明,胡铭曾;一种支持多线索快速切换的上下文机制:多上下文并行机制[J];计算机研究与发展;1997年S1期
相关硕士学位论文 前1条
1 王海洋;适于频繁上下文切换的TG-share分支预测器[D];中国科学院研究生院（计算技术研究所）;2002年
，本文编号：2126629
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2126629.html
